Bug #4925

Infinite recursion allowed in rescue clause

Added by Jonathan Dance almost 3 years ago. Updated almost 3 years ago.

[ruby-core:37346]
Status:Rejected
Priority:Normal
Assignee:Koichi Sasada
Category:-
Target version:1.9.3
ruby -v:ruby 1.9.3dev (2011-06-24 trunk 32222) [i686-linux] Backport:

Description

The issue pertains to the following example:

def a
b
rescue NameError
a
end
a

The type of exception raised and rescued is not relevant; the bug occurs with any exception type.

Expected behavior: A SystemStackError should be raised.

Current behavior: The Ruby process will consume cpu and memory until killed.

This is a regression in 1.9.x; the bug does not occur in 1.8-head. The bug is reproducible on all versions of 1.9 I have tested:

ruby 1.9.2p180 (2011-02-18 revision 30909) [i686-linux]
ruby 1.9.2p274 (2011-06-06 revision 31932) [i686-linux]
ruby 1.9.3dev (2011-06-24 trunk 32222) [i686-linux]

History

#1 Updated by Yui NARUSE almost 3 years ago

  • Status changed from Open to Assigned
  • Assignee set to Koichi Sasada

#2 Updated by Hiroshi Nakamura almost 3 years ago

  • Target version set to 1.9.3

#3 Updated by Yusuke Endoh almost 3 years ago

  • Status changed from Assigned to Rejected

Hello,

Expected behavior: A SystemStackError should be raised.

Current behavior: The Ruby process will consume cpu and memory until killed.

This is a regression in 1.9.x

No, unfortunately. This is caused by an intended spec change.

The root spec change is that 1.9 retains the full backtrace.
The change itself is useful, I think.

$ cat t.rb
def b
p caller
end

def a(n)
if n == 0
b
else
a(n-1)
end
end

a(10)

$ /usr/bin/ruby -v t.rb
ruby 1.8.7 (2010-08-16 patchlevel 302) [i686-linux]
["t.rb:7:in a'", "t.rb:9:ina'", "t.rb:13"]

$ ruby -v t.rb
ruby 1.9.2p180 (2011-02-18 revision 30909) [i686-linux]
["t.rb:7:in a'", "t.rb:9:ina'", "t.rb:9:in a'", "t.rb:9:ina'", "t.rb:9:in a'", "t.rb:9:ina'", "t.rb:9:in a'", "t.rb:9:ina'", "t.rb:9:in a'", "t.rb:9:ina'", "t.rb:9:in a'", "t.rb:13:in'"]

But this change impacts your code maybe unintentionally.

Whenever NameError occurs, the exception object creates a backtrace.
In 1.9, the time is O(n) where n is the length of stack.
So, your code runs in O(n2) until the stack overflows.

In addition, all backtrace objects are not collected because the
exception object can be accessible as "$!" in the rescue clause.
So, your code also uses memory in O(n2), which will cause thrashing.

You can confirm this behavior by clearing backtrace manually:

$ cat t.rb
def a
b
rescue NameError
$!.backtrace.clear
a
end

$ time ruby -v t.rb
ruby 1.9.2p180 (2011-02-18 revision 30909) [i686-linux]
t.rb:2: stack level too deep (SystemStackError)

real 0m22.361s
user 0m22.185s
sys 0m0.084s

The code works, but is still slow. I think that it runs in O(n2).

Anyway, this is not a bug, very unfortunately. Closing.

This may be solved by tail call optimization, but it should be
discussed in another thread. I personally dislike the optimization
in Ruby because it makes backtrace hard to understand.

Yusuke Endoh mame@tsg.ne.jp

Also available in: Atom PDF